Train the vision api with custom tags/domain specific models
I would like the ability to train the API with custom tags.
For instance, I would like to feed in pictures of cat breeds to 'train' the AI, then receive the breed value on an AI request.
We now have this type of capability using our Custom Vision Service. You can find further details to get started on the Custom Vision Service overview page.
Custom Vision Service is a tool for building custom image classifiers, and for making them better over time. For example, if you want a tool that could identify images of “Daisies”, “Daffodils”, and “Dahlias”, you could train a classifier to do that. You do so by providing Custom Vision Service with images for each tag you want to recognize.
I'd second being able to do our own training. Sans this, there is no real way to differentiate our own product
Stephan Heinke, MD commented
Categorisation of images based on some domain specific model is the main job of people doing visual medical diagnostics ( radiology, ultrasound ). It would be a very nice feature of MCS to allow some AI to support them.
Chris Sowers commented
I think providing a list of domains for MS to add is missing the point of the request a bit - it would be nice if customers could train their own custom models based on their own images - like other services such as Watson Visual Recognition or Clarifai.
Facial recognition matching against an arbitrary set (is it me ?)
Images of plants tagged eatable (bool)
Images of foods tagged by calorie count (float)
Watson does this I think.. Though retraining/updating is tedious.
Dhiraj Gupta commented
I also need a way to create a group of objects to identify them later, Just like a person Group in face API. When can we expect this functionality.
Chris Sowers commented
+1 for this. Very high interest in training domain specific models via these APIs or Azure ML..
Thanks Ryan & Benny. We do not have the functionality available today, but it's extremely helpful to know that you're interested in it. This is actually already tracked by a different request in the forum, so I'm going to merge these two articles together.
Appreciate the feedback!
Benny Olveng commented
I would like to use the Computer Vision API with my own set of medical pictures to train the neuralnet . Is this possible ? I guess this is why the thread is started - its not possible yet ? ( or have I misunderstood and should start another thread ?)
Ryan Moore commented
Thanks for the speedy reply. We are interested in using the Computer Vision API with our own domain specific model. I understand that the 'celebrities' model is the only one available presently, though it looks like the Fetch! app from the Garage also uses a custom model. The ability to train a custom model through the Computer Vision API, or work with Microsoft to develop one, would be super neat.
Thanks for reaching out! Is there a specific API you're interested in for training a domain specific model?